Data Science Asked by Manimaran Subramanian on July 30, 2021
In one of course Andrew Ng mentioned that model (Machine learning) error can’t outperform Bayes / human level performance and hence the bias can’t avoided.. (Unavoidable bias)
How do we determine / estimate Bayes error or human level error so that we could stop optimizing the model for reducing bias further?
Examples / concepts for classification problem would be appreciated…
Bayes error and human error are two different concepts. Bayes error is the theoretical lowest error possible on a task, there can be no lower error rate. Human error is the empirical lowest error that a human can perform.
Since Bayes error is theoretical, for most non-trivial tasks Bayes error must be estimated based on domain knowledge.
Human error can be found by having human perform the task.
A popular example is ImageNet. Current machine learning systems are better than humans error (top-5 error rate of ~4-5%) and rapidly approach Bayes error rate (top-5 error rate of <3%).
Answered by Brian Spiering on July 30, 2021
Get help from others!
Recent Questions
Recent Answers
© 2024 TransWikia.com. All rights reserved. Sites we Love: PCI Database, UKBizDB, Menu Kuliner, Sharing RPP